Third Bridge专家称AI正推动广告代理与广告技术平台商业模式重构

[[00:01:34]]
Let's talk about AI's impact on the broader advertising agency and holdco-level space. How would you describe the core North American advertising agency business today, compared to 3-5 years ago before the AI acceleration really happened? There were similar large tech risks then, but I think this is fundamentally different. Could you discuss the differences between then and now?

[[00:02:11]]
I think our world is changing a lot, obviously. As we look at how today we're looking to monetise and live on. I think the biggest thing is our commercial models are fundamentally shifting and changing. We're aggressively trying to find a way to move out of a time-and-material model. As we create new layers of technology, and now the holdcos are obviously leaning in on this platform world, how are we doing so in a way where we're not just cutting FTEs and cutting time and then not being compensated for it. What does that look like? How can we do it? Is it an outcomes-based model? Is it a token-based model? What does that look like? That's, I think, what the entire industry is really struggling with to find and figure out. That's the big change in the commercial model and that the time-and-material model is dead. Obviously, a lot of the revenue is still tied to it, but we can just see that it's a ticking time bomb, and we got to figure that out. The other thing I would say is I think that size and scale largely doesn't matter anymore. With that is going to come a new wave of start-ups, which I think we're starting to see that are tech-first, AI-native start-ups where brands and groups can go in and integrate into them a lot easier. They are much smaller. They're much more nimble. They're able to outmanoeuvre the holdcos.

Then last, last, I would say, just like the broad industry as a whole has really done a bit of a shift. If you look at the LUMAscape end-to-end from brand through to publisher, the adtech layer has shifted. Everyone is trying to move further upstream in the demand side, where you have book/bill/pay systems trying to move into the SSP space. You have SSPs moving into DSPs, and also, somewhat moving into the agency space through the platforms they've created. Then you also have DSPs going direct to brand where they've got direct integrations into publishers, so cutting out the agency and the SSP. It's those three big things that inside of the last, I don't know, 6-8 months have really started to show, and I think are just going to ramp up even further as we progress, I think, in the next six months, we'll see a lot of that really come through.

[[00:05:03]]
Let's think about convergence when it comes to the adtech landscape as a whole. How do you ultimately see that playing out, and what will be the read-through down to the holdco level? Are they going to go through a similar convergence, where nobody is really differentiated and, at the end of the day, everybody is competing along the same lines, even down to the agency level? How are you thinking about broad media buying convergence as a whole, including the buyer and holdco sides and down to the adtech platforms as well?

[[00:05:51]]
Ultimately, I don't know how long this is going to take, obviously, but I think ultimately, this is going to result in fewer partners that handle more volume at the end of the day, I think that there's going to be more end-to-end platforms across it all. For example, some of the leading SSPs today are already able to compete with a lot of the holdco offering, primarily on the media side, but just broad strokes. With that, they're going to create essentially an end-to-end platform that can compete with the likes of the holdcos. When we get to that state and they decide to bite off the hand that feeds them, quote, unquote, the first mover that does that will then result in more of a convergence to your point, where we then end up with just end-to-end platforms that are differentiated on three things.

They're going to be differentiated on their segments. Think of it as three Ss where you have your segments or your data or identity spine, if you have one, your science, which is going to be all the decisioning data that you can put into an AI-based model because the AI and the model itself isn't going to really do anything for you. It's going to be highly commoditised and at parity. Then the last one is going to be your scale. When I say scale, I mean more of like what media rates can you get for the partners that come into this ecosystem that you're going to buy through. How do we look at getting the cheapest and the best ROI for the brands that come on the platform? It's going to be those things. I think that there'll be a select number of them that are in the ecosystem that go end to end, but you won't have any of this adtech hopping that's needed.

[[00:07:54]]
You mentioned the idea of there not being much advantage left in buying at scale. How do we reconcile that with what you said about scale being a potential differentiator at the holdco level?

[[00:08:19]]
There's scale in that the more dollars that flow through your pipes, you can achieve economies of scale where you're going to be able to get preferred rates and deal terms. Historically, as we've talked about scale as in, "We're in 92 offices globally, and we're in every market," you don't need that anymore. That's actually going to be a lagging factor that's going to slow down and allow other people to come outmanoeuvre you and flank you. Because of the way that these groups are set up, a lot of senior executives are just trying to hold on to power, I get that, and with that comes their pay cheque. If they start to relinquish FTEs, if they start to relinquish offices, they no longer need to have a job. This then is very disruptive as the ecosystem is being disrupted. Think of scale in two ways, one, a monetary scale, which will be a differentiator and who puts the most money through their pipes. Think of the physical scale as a negative asset.

[[00:09:32]]
How should we think about the end-to-end platforms? You mentioned that we're going to live in a world where you're looking at probably 10 end-to-end platforms. What do you think ultimately drives the winners there? This goes back to the AI conversation. Is it just the ability to leverage AI to build these platforms faster and truly integrate all of the disparate parts? Is it just a race? Are there certain capabilities that people are going to be able to nail down and others aren't? What ultimately leads to the winners from a tech perspective?

[[00:10:25]]
It's going to be based on who moves quickest and who moves best, but I think that the science and the decisioning, there will be some layer of like, "We can drive better performance than you because we're sitting on this wealth of data. We have really good historical performance indicators, which then feeds into a predictive model that allows our performance to be a couple of basis points higher than the other player. When that's in play, you want to put your dollars through our pipes because of the ROI on a percentage basis could be massive for your brand." I think that's on a technology basis, what's going to result in a clearly-defined winner. I do believe that will still be a thing. As we look at adtech more broadly, no one wants to really ever talk about this, but we look at QPS and we start to get really in the nitty gritty of how these bidders work and what drives real performance on them where there's going to be latency caps and all these kind of components, the adtech group may be ahead of the holdcos on being true technology players and understanding that best. I think that that will be a determining factor as well.

[[00:11:46]]
Thinking about the convergence question, do you think there's a world in which there's no differentiation between a holdco and a platform, or is that an extrapolation too far?

[[00:12:09]]
No, it's not. If you look at the LUMAscape today, and I don't know how familiar you are with that as a concept, but there's buckets of brands of logos that sit inside of these circles. They show the hop. There's an agency moves to a DSP, which moves to an exchange to an SSP to a publisher. Tomorrow, there's just going to be… Imagine there were just five lines running straight across it and one was an Omnicom line, one was the Magnite line, one was a PubMatic line. There were just five master platforms that just sat over the entire LUMAscape and that becomes, basically, the marketing ecosystem in a nutshell.

[[00:12:57]]
The performance of a player such as The Trade Desk has been quite dismal lately – there have been a lot of concerns around AI with these players. If adtech players are so far ahead, how do we reconcile this? The Trade Desk, for example, was the darling of the space perhaps one year ago. What are your thoughts there?

[[00:13:34]]
What Jeff Green did there was he quickly realised essentially that he juiced his margin and did a good job, and he basically squeezed the agency orange for all that it was worth. Then he said, "Okay, I need to find new sources of revenue growth. Where am I going to get that from? I need to move further to the brand, take the agency cut out of the equation." When he started to do that, he then got into some kind of hot water from both the agencies, but also from the SSPs because in order to do that, he went direct to the publisher. That's what actually happened. The reason why that could end up being successful is he's now, in effect, going to be the first mover to try to compete direct to a holdco. There's been some light mediation between Publicis and between The Trade Desk on some of that flare-up, fine. The reality is he's got the pipes direct to the publishers in the works already working. He's already started this track down direct to brand. It will be an end-to-end platform where they can do it all in the near future is my guess.

[[00:14:51]]
Is there any advantage to being close to the publisher? Do those relationships matter? Is there any advantage or alpha to be generated from, in particular, working better with a given publisher? Amazon, for example, has its DSP [demand-side platform], but it's also a publisher. How are you thinking about that?

[[00:15:25]]
I think Amazon is an interesting one, but how I'm thinking about it is I'm not sure what happens to the Amazon DSP long term, but I don't think that Amazon can continue to operate a neutral, let's call it, DSP and operate a very large growing ad platform, an Amazon Ads, just to say that. I think that you're going to need to have these neutral, quote, unquote, partners of which The Trade Desk could and will be one where they can buy the Amazon Ads inventory. I don't see a world where you have the Amazon DSP as you're supposed to have these master platforms or you have the Amazon DSP integrating into the Google ecosystem. That's not going to work, is my guess. I think, honestly, the same goes for Google and some of their buying mechanics or whatever we want to call them today, as that world is quickly evolving, but I think the same applies there as well. To answer your question, I think that working with a specific publisher, no, I don't think there's any real benefit. The only benefit would be that you can make very large commits to them in a principal media-type capacity, and they're going to give you preferred rates, but they're also maybe going to give you better access to roadmaps, so on and so forth on their sell-side tech.

[[00:16:50]]
Are we talking about a conflict of interest? You mentioned the neutrality. Why hasn't that happened yet? For the most part, everybody is using DV360 through Google. This is only a publisher through its social apps. Why has that not been a factor yet if it's going to be down the line?

[[00:17:34]]
I think as you start to look at how the ecosystem changes, it's going to become a smaller part of the revenue that would then distract and create enterprise issues where they'll just shut it down because they won't be worth the headache and won't be worth the large deals that they then are not allowed to partake in. I think that the cost to compute on a lot of this stuff when you look at the large cloud deals, when you look at the tokenisation and who owns the models and who's going to monetise the model, that's going to become a larger part of these businesses and the ad revenue is going to become pretty small where Amazon is just going to say, "We don't need to have an O&O DSP anymore. We're charging 1%. We did it, ultimately, to push more dollars to our ads business. We've done a good job. That's a grown-up business unit now. It's pretty established. It runs good performance. We don't need the DSP anymore, shut the DSP down." I can see that happening.

The other thing is he who owns the decisioning right is going to be a huge thing in the future. If you look at how holdco actually operates today, if you actually think of it in a simplest form, it's ridiculous. We have allowed ourselves to work direct to client. We pitch a campaign idea to a client and then we take that, the client leaves the room and we swivel chair into someone else's technology, we keystroke into that technology and they run all the decisioning in their black box, and then they don't give us any of the log-level data that we can then read out on it and see how it actually performed up against comps and up against other platforms. That will change. You'll have five master platforms that control their own decisioning rights and the DSP will not. The holdcos are going to have to create or evolve into a DSP-light mechanism. There's a lot of ways that you can talk about that happening and working through, but that's going to have to happen. I think when that happens, something like an Amazon DSP won't be a thing anymore.

[[00:20:04]]
Doesn't what you're outlining undermine the positioning of the holdcos today, given that they maybe aren't getting the outcomes data at the level that they should be? So much of this conversation is going to go back to data. You mentioned predictive models and how getting a few basis points of different in terms of performance is going to be a differentiator. Wouldn't that theoretically put the holdcos even further behind the eight-ball that we're giving them credit for?

[[00:20:55]]
Yes, absolutely. Yes, in a huge way. What happens when The Trade Desk just says hypothetically, "Omnicom, you're not going to be able to spend any money through our platform anymore, we're going direct to your clients because we know you don't have many decisioning and we have a platform that can do all the things that you're going to do." The holdco needs to have an answer for that. Now there's a lot of things that they could look to do and I think this will all happen pretty quick is obviously, there's M&A, obviously, there's in-house build like this joint venture partnerships. There's a lot of ways that that can be addressed as well and will be addressed by the winners. Yes, I think that's correct, how you're seeing it.

[[00:21:40]]
For around five years now, Epsilon has been somewhat of a holy grail in terms of holdcos, with people saying, "It's a great data spine" and all of that and spurring others to pursue similar capabilities. Do you think that helps reach the point faster? If we think about this as an example, does that strengthen the value of the data spine?

[[00:22:23]]
Let's talk about that. They have a data spine A, which is helpful. What you then need for that data spine to work is you need a data activation layer. You need to have something that is in the bidstream that can be activated on down to the individual ID layer in order for you to really see the value and be able to do a DSP-like object off of the Epsilon deal. The reason that is important to say that is Publicis just went out and then acquired LiveRamp. They didn't acquire it to then be a neutral third party where they continue to take other holdco dollars on the RampID. They acquired it to create a walled garden-like object where they have their own enterprise strength DSP, but they also have their own O&O data graph in Epsilon and they combine these pieces together. My hypothesis there is it, quickly, is no longer a neutral asset. You can see those of… There's talk already in the market publicly in some investor forums, JPMorgan, you can figure that out yourself, but it's public where other holdco CEOs are saying, "My group is going to be off the RampID by X date." I think that's fair. I think that's real. It's no longer going to be a neutral partner, and it's going to be what powers the Publicis' decisioning engine that they're building inside of CoreAI. That's a guess. I don't know if that's a true…

[[00:23:58]]
You mentioned how things are going to look increasingly walled garden-esque. We've seen some walled gardens come down and others not so much. Going forwards, do you see that having any effects one way or the other, where there will be more and more walled gardens? Is everything going to be a walled garden? If we're talking about the few platforms that remain, and each one operating independently, does that then create an opening in the market for somebody else to then integrate all of them? These are going to be closed ecosystems. Is that what you're getting at?

[[00:25:14]]
They're going to be closed ecosystems that connect into other closed ecosystems meaning Publicis is going to be a closed ecosystem. If you want to partake in what they're offering, you have to go through their walls, you have to use their pipes, so on and so forth, outcomes-based models, token-based models, etc, all the things we said before. They will then connect into these walled gardens. They'll have their RampID in the walled garden and they'll be able to get both injecting that ID down to the deal ID level, where they can run every campaign through their pipes down to the individual, where then they'll then have the ability through the walled gardens, they'll have deals brokered where they can also retrieve that exhaust level data through all the walled gardens in a master clean room-type object where they can then measure, report in near real time on it, and then get to this always-on optimisation holy grail, so on and so forth. That is how it'll go.

[[00:26:22]]
What you're saying sounds great – it's just a case of changing the monetisation model – but I'm sure that's a lot easier said than done. Could you outline the challenges around updating this monetisation model? You alluded to some of the legacy practices here, the hesitancy to change and how this is going to be a big disruption overall.

[[00:27:11]]
You have a couple of things here in play. There are a lot of legacy businesses that are highly profitable where you have time-and-materials, you have FTEs on them, they're massive in terms of headcount. That's one thing. From the holdco perspective, you have cost-out work that's going on, where groups are under pressure to drive high margin numbers. Where then do you find the fat to cut the cost? There's talks on, "You can cut some of the billable FTE people," but then they're like, "Okay, but you're going to destroy the top line." There's a lot of challenges in that just baseline in the current make-up. I think that in the future, it will actually be less of a challenge, just more so in that from a net-new business perspective. I think once you get a couple of these under your belt, the others will follow and you'll be able to go proactively and actually go in and change terms against them because the brands are going to look for this as well, where you have more performance-oriented commercial models that you're now able to actually pay off once Google Assist is able to do what I just stated through the LiveRamp deal. I think that there's, obviously, challenges, but there's a path forward through those challenges where that's not as much of a challenge.

[[00:29:03]]
Has anybody got started with what we're mentioning? You mentioned the pressures. Is there a model out there that's easier to transition? It goes back to some of the tech platforms. How are you thinking about who will ultimately rip the band-aid off first there, given that you're likely to take a hit?

[[00:29:40]]
I think that The Trade Desk oddly enough is the furthest ahead on this. I was surprised that they walked back some of… They made peace with Publicis. I think that ultimately, they're in an interesting position where they had an opportunity to rip the band-aid off and go for it, which they then didn't do. I think that they're in an interesting place and could do it. Then I think that the SSPs are also in a very interesting place, too, where some of them had made good interesting acquisitions of essentially DSPs where it allowed them to then create these platforms that essentially you're putting a DSP on top of an SSP, you're putting an agentic layer on top of both that inventory by MCP. Once you can do that, there's a lot that you can do, and they're sitting on all the performance data as well. I think a lot that can happen there, too.

[[00:30:49]]
You mentioned the agentic layer and putting it on top. How realistically can somebody do that? Are you saying that, if you wanted to rip the band-aid off, we could be getting these fully end-to-end platforms running on an agentic layer by 2027? Is that crazy to say? How are you assessing that?

[[00:31:31]]
Yes, the answer is if you look up what Magnite and PubMatic are doing… PubMatic is very open in their PR on what they're doing on this. This is public. They've got their AgenticOS. They're running these campaigns largely today through small independents, but that's because the indies are able to move quicker or are more interested in running press around it. There's constraints on running press for the holdcos because if they run press around it, to my earlier point, they haven't really worked out the commercial model against it yet. That's an interesting thing. The SSPs today are like they're there. They've productised the offering and they're there. The clients that they would need to be going to are going to be less on the managed service side that want those big teams, that want that big account lead. It's going to be more of the clients that are performance-oriented. They have in-house shops and they're maybe a little leaner just in how they operate globally and in select markets. That's very easy for them to do. It's just the classic thing of, if they make that move and they start going direct to brands in a way where they're going to have an O&O relationship with them, what does that do then to the holdcos and what's the retaliation, if any?

[[00:33:03]]
You mentioned what Magnite and PubMatic are doing with their AgenticOS. With so much of the AI conversation, we should probably think about what's spin and what isn't. What is that enabling the companies to do today? Does this still feed the ultimate decoupling from the holdco? Is that the larger thing that we're getting at?

[[00:33:48]]
What it allows them to do today is what they're showing, it allows them to do and what they're doing is it allows them to expose their inventory to a buy agent, which basically means that through an agentic layer, if you have a buy agent inside of an agency or if a brand has a buy agent that they own themselves, you can buy media through them agentically. That's at a macro level. At a very micro level, it means that you can run that against multiple SSPs if you were going to lean on them as a historical aggregator, and you could start to assess in near lifetime with an agent that you have sitting on top of it, different inventory availability against different audiences. Then you can run a decisioning agent just based on all this decisioning modelling data that I'm talking about where you can then activate through them. What it's allowing them to do is cut out the DSP because you're able to put their DSP plus their SSP on top of their inventory and activate and do it in a way where the MCP just, basically, collects all of the activation information that you would need and you can activate from inside of your own wall, your own buy agent or you can create the line items inside of their platforms all agentically and then you can go into their legacy platforms, their legacy UIs and you could QA it there.

[[00:35:19]]
We hear about other agentic shopping down the line. How close are we to true agentic buying? We're talking about agents and no longer needing the hands-on keys for the programmatic buys anymore. When do you think that will happen? Do you think that will ultimately lead to more or less total ad spend? What are the ramifications of this?

[[00:35:54]]
This one, I think, is a little further out than maybe some people are selling at the moment, that's my two cents. I think that it's coming. I think it's real. I think it's a thing. I have yet to see to the individual persona where you have a personal agent or a personal buy agent. I haven't seen that in the market. I've seen some groups that are close. I think that you would need to have a personal agent that feeds and reads off of all of the data that you have that you give it access to, and it would then need to be directly connected into marketplaces. I think that's a little further off. I think then that does result in lower ad spend, but I think it results in differing in how media has been transacted, meaning instead of looking at things on a CPM basis through OpenAI, how they sell the intent that comes through that system, just looking at how much lower in the funnel you can get from the jump of querying into an LLM vs legacy search.

I think that will just become a lot different and then (audio cuts out 37.17) and you can connect into them your personal agent, and because you just jumped right into it, the brand then is going to pay a premium for that sale, if you will, much more than we would see in today's world. I don't know if we would call that advertising. I don't know if we would call that better commercialisation deals of data. I don't know what you would call that, but I think that the down to the intent level will be much more expensive than it has been historically.

[[00:37:49]]
Would the publishers capture the premium? Will the premium be captured by the marketplace or ultimately, in this case, whichever the DSP wins?

[[00:38:04]]
It's also assuming that the publisher is going to be the one that even gets a cut of the attribution of your agents saw that you viewed this now. We could really talk about that one quite deep technically. I'm talking about agentic buying where you have a… Maybe we're looking at it through different lenses now I'm thinking about what your question is. Are you talking about agentic buying on a personal level, a consumer level or are you talking about agentic buying on…

[[00:38:34]]
You mentioned how there might be slightly lower ad spend, but also, if we're going to be able to capture that intent data, you mentioned the premium factor there. Wouldn't the publisher eventually capture that?

[[00:39:09]]
The way it's going to ultimately go is you're going to have a buy agent and then you're going to have a sell agent. One thing that there's going to be different tranches of this. When we look how it's going to converge, you're going to have five platforms, just say, hypothetically, that connect the marketer to the publisher, and it handles all that transaction. The way that that's going to go on the first wave is you're going to still be connecting through mass aggregators because the sell agent technology is not there yet. There's a couple of groups that are building pretty interesting buy agents. You look at Scope3, you look at a group called Kovva. There's interesting groups, too, that are also building a sell agent technology.

You have Swivel AI and you have a group called Bedrock, which is labelled as a DSP, but is also building sell agent tech where the sell agent guys are going direct to these big publishers and they're saying, "Hey, we can expose your inventory via a software-type play. You give us access to your database of your inventory and how we would transact it. We'll expose it via MCP, so buy agents can buy it." Until that reaches critical mass, you're going to have to go through aggregators, which are the SSPs today. In a future state, Disney will have its owned-and-operated sell agent, which will sit against some master protocol so that anyone can access it and anyone can buy Disney ad inventory. Smaller publishers may not have their own O&O sell agent, but they will have a software subscription-type thing that they pay for where they give a cut of it to the group that exposes the inventory. That's going to be the modern SSP, but that's a little ways out.

[[00:41:05]]
Could you outline the scenario where the the holdco models ultimately catch up? What leads us there? We've talked about who's well-positioned here. You mentioned what The Trade Desk did, even though it walked it back, to some extent. You also pointed to the SSPs [supply-side platforms]. What's the path for the holdcos ultimately platformising and winning this race?

[[00:41:44]]
How and when do the holdcos, how do they catch up? They catch up because they still control the lion's share of demand. They're in a position of power for now. Even if that starts to dwindle, I think there's enough time for select few of them to catch up. I think little groups like a Dentsu, so on and so forth, they're going to really struggle and they will never catch up, and they'll just be sold for piece and parts.

Ultimately, it goes back to, if you look at the three things I talked about, if you look at Publicis, if you look at Omnicom, they both have an identity graph and an identity layer. They both have an ID-level currency in Real ID and in RampID. They both have a wealth of historical decisioning data that they're sitting on, and they both have the economies of scale at the moment where they can get really good rates. They're going to struggle on, "We have to close all of these offices. We have to lay off all these people." The holdco of the future is not going to be 100,000 heads. It's going to be 10,000 heads. That's just a fact. There's going to be a lot of pain and a lot of struggles there to get to that. I think because they control the pocketbook of the marketer overwhelmingly still today, I think it buys them time, which allows them to catch up.

[[00:43:37]]
I read an interesting article about how the holdcos are essentially saying, "We're actually running ahead on AI. It's the buyers who are not ready." What would you say in response to that? Is there some truth to the idea that the buyers are ultimately holding back part of this big transition here, or do you think that might just be spin?

[[00:44:10]]
You're saying, will buyers, meaning, will brands want to go slower on some of this? Is that what you're asking?

[[00:44:23]]
Yes, essentially.

[[00:44:29]]
Yes, a couple of things. I think some brands are going to want to go fast. Some brands will go fast and they'll pave the way for others to then move faster. I think that's really real, and that will happen. That's very true. There's a lot of hesitancy on what does this mean, how does this work, what do we do. I think that's a thing for sure. I also think that some of the walled gardens are going to move slow. None of them have active live MCPs where they're allowing exposure to their information because they want to really monitor it and they're very happy with their current API today. They're all in the same place on that at the moment. I think you're going to have hesitancy from the brands, but also from a lot of the large walled garden players to… They don't want this always-on optimisation. They want the dollars to continue to come in as they have because I think they know that it's more advantageous to them if that same model stays in place.

[[00:45:36]]
How could it be more advantageous for certain brands, if that's what you're saying?

[[00:45:44]]
Because today, there's a lot of marketers that just push… It's a safe bet to say, "Put my money in Google and Meta." We've assumed that that's actually the best place for the dollars and the best ROI. I think that's a hypothesis that we think that's the case. If we get into an always-on optimisation model and something starts to perform better on Snap or on the Open Web or on CPV, those dollars start to rotate in lifetime out of those platforms. That creates a hell of an issue for them. They don't want to allow that access to an always-on component. That's my guess.

[[00:46:28]]
We chatted in a previous Interview [see Stagwell – Growth Outlook & Technology Update – 23 January 2026] about the collapse of the media supply chain. In terms of the margin or premium that's being paid now for all the disparate intermediaries, do you think that's a legacy component? Is that down to employee-level inertia here? Can it be described as that? It seems it would still be advantageous, given all of the margin that's left on the table for other intermediaries to accelerate this.

[[00:47:10]]
Yes, but there's a lot of margin, but a lot of the supply-side publishers, I think, they view it more as a risk than they do a plus. Are they going to actually see more working media dollars and are more of those dollars going to actually flow into their ecosystem, that's the bet that they would have to make on the strength of their inventory and their actual targeting models and so on and so forth. You're then going to start to see real performance metrics and you're going to get exposed if you're not actually where you say you are.

[[00:47:52]]
We've seen a lot of talk around how AI is leading to efficiencies and margin improvements by some of the holdcos. Where do you think that's ultimately playing out today? Are you seeing areas where the AI is driving legitimate efficiencies today?

[[00:48:24]]
AI is driving efficiencies today? Honestly, no. No, I wouldn't say that there's any efficiencies today. I would say that we can quickly see a world where those efficiencies are going to be unlocked through this agentic media buying process, where you're going to need fewer people to buy and transact agentically through the media world, and you're going to be able to get to much better speed to market, speed to value. It's right around the corner. We can see it, we can feel it, we can taste it, and it's close.

[[00:49:06]]
Going back to what we said about where the holdcos may shake out, assuming that they do all catch up, you mentioned that there are 10 end-to-end platforms, some of the biggest of which are players such as Publicis. Alongside players such as The Trade Desk and some SSPs, are they fundamentally small? We can look at this from a headcount perspective, but how would we then size the revenue opportunity there? Are they going to be gaining so much in terms of the consolidation of media buying to the limited platforms that this will actually end up being a net positive?

[[00:49:59]]
For those groups, yes, it's massive upside for some of these groups, huge for them. Look at the size of The Trade Desk revenue today. If that alters because they bite the hand off that feeds them and the SSPs play this right, The Trade Desk loses 20-30% of its revenue, where does that revenue go? In the short term, it goes to Magnite, it goes to PubMatic because the holdcos don't have decisioning and the build yet. It's a massive upside opportunity for these guys.

[[00:50:29]]
Margins have been under pressure at the holdco level. I'm not saying that The Trade Desk and the SSPs are going to have the same issue. At the end of the day, you're still functioning, in a way, as an intermediary. Assuming that this will be a bit of a differentiator, is there a healthy margin there at the end of the day, or will it be the same and be a case of, "We're working off a bigger figure, but we're still going to be working on the same 10% EBITDA margins at the end of the day"? Is that how you see it playing out? Wouldn't any excess margin eventually be captured back by publishers?

[[00:51:26]]
Yes, it depends. That's what the big struggles with when I say we don't know what the commercial model looks like today. Who actually captures that margin and for how long, that's up for debate. I think it's going to be really interesting to see. I think that The Trade Desk has proved that you can run an adtech company at a very profitable high margin and for a very lengthy period of time. The reality is, yes, they're a relatively new company, but they've run at very nice growing margins for a very long period of time. They're a public company. Everyone knows what their margins are. We all know what's going on. We all know it's not transparent. They have a talk track that says otherwise, but just look at the numbers, it doesn't take a rocket scientist to figure out what's going on. I think they proved that you can do that if you're smart with how you set up your model.

[[00:52:27]]
What do you think is the biggest risk in everything we're discussing? Platforms are going to go by the wayside. Perhaps it goes back to this commercial model. Obviously, we're not there yet, but everybody talks about the software platforms being completely replaced by vibe-coded AI solutions in-house. Is there a world where this could lead to more in-housing and even further disintermediation? Is that a realistic possibility? How do you think that will play out?

[[00:53:06]]
There will be more in-housing, but the question isn't, will there be more in-housing, it will be who owns the technology that powers the in-housing. The holdco tomorrow is also going to have a self-service application that feeds off of their same platform. Imagine a world where today, I don't know, you're Best Buy and you manage USD 1bn of in-house performance media for Best Buy. You in-house and you have, whatever it is, 800 people that do that for you today. I have no idea what the numbers are. I have no idea. A holdco could come to you and say, "Here's a layer of tech. You're going to get access to our audiences. You're going to get access to our decisioning. You're going to get access to our economies of scale. All we need from you is access to monetise that USD 1bn and you can take this tech for free." Then it's on the holdco to find a way through commits, through principle, through better decisioning, better deal terms to monetise that USD 1bn to make a profit on it. Best Buy then says, "Okay, great. We're going to get better performance. We're going to get lower rates, and we don't need to employ these 800 people anymore." Now a holdco, when I say holdco, I mean an end-to-end platform, Trade Desk, for example, could do that.

[[00:54:42]]
What do you think is the clear inflection to the next acceleration of the transition? This is right here on the horizon, although we're not there yet.

[[00:55:06]]
I think that the LiveRamp deal is going to close in, let's call it, January, February, we'll see how that goes. I think that's one of the bigger inflections because I think that we're going to start to see their positioning change in the market and people will react to that as they're the leader. I think that we got to watch The Trade Desk/the leading SSPs on how they want to play this and where they think the time is right to push the envelope. I don't know if there's necessarily an inflection. I think it is still a little bit of a waiting game. I also think that M&A is going to pick up. If a holdco finds themselves behind on decisioning, what exists in the market today that they can go out and acquire to help them catch up and bolt something into their existing system to get them caught back up, I think that world is not done at all. I think LiveRamp was the first of many.

[[00:56:14]]
Who else out there do you think would fulfil a similar role to LiveRamp?

[[00:56:28]]
That's just one niche piece that Publicis felt they needed. It doesn't mean that they couldn't go out and acquire an SSP. Why would they not go out and buy PubMatic or buy Magnite, for example? They have a DSP-type functionality. They have a data layer. They have an identity layer. Then they have a supply side layer. They have the pipes now into all the supply. Those kinds of moves, I think, are coming.

[[00:57:01]]
Would you like to highlight anything else around AI in the advertising agency space?

[[00:57:11]]
I think that the creative piece is an interesting one. I think it's highly commoditised, but I think that the holdcos today, it's still revenue and margin accretive for them, but it's going to become more and more of a drag. It's going to become an issue for them, where a lot of the adtech groups have either acquired, like Magnite acquired Streamr. They've either made acquisitions in that space to combat that or they're just going to build or leverage creative models. If you look at Publicis, if you look at Omnicom, you look at Havas, they have huge creative arms in the tens of thousands. That's going to take a while to unwind that.

[[00:58:02]]
Why hasn't that already been replaced today? Not from Publicis's perspective, but who's out there still buying creative when there are so many different alternatives, at a high margin, obviously?

[[00:58:26]]
Yes, brands still are. It's the reality, and they are margin-accretive. They're revenue- and margin-accretive. I think the size and the growth of them has declined. Does not think, it has definitely declined across the industry. There's different integrated models where the media offsets the decline in creative for those brands. There's a lot of ways that that's been skinned to allow that to continue to live on, but that's going to quickly erode. It's going to quickly erode, but then to quickly get out of it and then you're sitting on top of a bunch of office space, there's a lot of cost to that, whereas there's going to be adtech groups that are going to be in hyperscale mode that are 1,000-2,000 FTEs that can just outmanoeuvre these holdcos.

[[00:59:37]]
Thank you so much.

[[00:59:38]]
I love these as always, so thank you.

[[00:59:40]]
For sure. Pleasure is mine. Take care.

Transcription ends at 00:59:42 of the recorded material